KerasでステートフルRNNを使ったサンプルコード

keras_stateful_rnn.py

# -*- coding: utf-8 -*-
import numpy as np
import pandas as pd
from sklearn.preprocessing import StandardScaler
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, SimpleRNN, GRU, LSTM

#%% data preparation
df = pd.read_csv("osaka_temperature2009_2018.csv",
                 index_col=0, parse_dates=True)
df = df.interpolate(method="linear")

#df.plot()

ss = StandardScaler()
std = ss.fit_transform(df)
std = std.astype(np.float32)

#std.plot()

#%% data arranging
timesteps = 6
batch_size = timesteps

x = np.empty([len(std)-timesteps, timesteps], dtype=np.float32)
y = np.empty(len(std)-timesteps, dtype=np.float32)

for i in range(len(x)):
    x[i] = std[i:i+timesteps].T
    y[i] = std[i+timesteps]

data_len = batch_size*int(len(x)/batch_size)

x = x[:data_len].reshape(data_len,timesteps,-1)
y = y[:data_len].reshape(data_len,-1)


actfunc = "tanh"
N_EPOCH = 3

#%% stateless model
model = Sequential()
model.add(SimpleRNN(10, activation=actfunc, 
                    stateful=False,
                    input_shape=(timesteps, 1)))
model.add(Dense(10, activation=actfunc))
model.add(Dense(1))

model.compile(optimizer='RMSprop', loss='mean_squared_error')

history = model.fit(x, y, epochs=N_EPOCH, batch_size=batch_size,
                    verbose=1, shuffle=False)


#%% stateful model
model = Sequential()
model.add(SimpleRNN(10, activation=actfunc, 
                    stateful=True,
                    input_shape=(timesteps, 1),
                    batch_size=batch_size))
model.add(Dense(10, activation=actfunc))
model.add(Dense(1))

model.compile(optimizer='RMSprop', loss='mean_squared_error')

for i in range(N_EPOCH):
    history = model.fit(x, y, epochs=1, batch_size=batch_size, verbose=1, shuffle=False)
    model.reset_states()